data: add qwen3.7-plus & mimo-v2.5 JEE Advanced 2026 results; rescore 2026

Fix JEE Advanced 2026 numerical answer-key ranges; add re-scored results

config: add Gemini 3.1 Flash Lite

config: add Mistral Medium 3.5, Kimi K2.5, Gemma 4 31B

config: add x-ai/grok-4.3 to model list

config: add openai/gpt-5.4 to model list

config: add anthropic/claude-haiku-4.5 to model list

feat: extend reasoning_effort to qwen/* and z-ai/*; add Qwen3-VL and GLM-4.6V

feat: enable thinking effort for Google Gemini models (default high)

config: remove o3, keep gpt-5.5

config: increase max_concurrent_requests to 8

config: raise max_tokens to 32000 and request_timeout to 600s for thinking models

feat: concurrent API calls via ThreadPoolExecutor (default 4 workers)

docs: clarify max_tokens comment — high value is intentional for thinking models

Update models to May 2026 releases

Add reproducibility controls, leaderboard, and contamination disclosure

Move metadata.jsonl into images/ so ImageFolder picks up metadata

Move metadata.jsonl to repo root for HuggingFace image resolution

Replace dataset loading script with direct JSONL loading

Fix 7 medium-priority issues: resume, timeout, logging, prompts, dead code

With gpt-5 results

O3 results

Add Gemini Flash model to benchmark config and save results

feat: Add Claude Sonnet 4 to OpenRouter benchmark models

JEE_ADVANCE to JEE_ADVANCED

Updated benchmark_config.yaml

With more clear summary report and reattempt mechanism

First working version